The Grifters, Chapter 3 - Election Prediction

Election prediction models assign precise odds with scientific confidence. Yet the same models that called Clinton at 72% in 2016 and Biden at 90% in 2020 missed actual margins by enormous amounts. The gap between what these models claim to know and what they reliably predict reveals something uncomfortable about their actual purpose.

• The models fail at the thing they're explicitly designed to measure. When polls capture strategic responses instead of genuine voter intent, and when voting behavior shifts in ways surveys can't detect, the mathematical framework doesn't just underperform. It converts embedded errors into false certainty.
• Polls stopped functioning as independent signals long ago. Both pollsters and respondents now operate with the knowledge that their numbers will be used as ammunition in political games. This shared understanding changes how people answer questions. It's not deception. It's rational strategic behavior embedded in the system itself.
• The financial incentive structure has nothing to do with accuracy. Media companies acquired prediction platforms to create months of daily score updates, viral debates, engagement spikes. That's the actual product being sold. The election forecast is the wrapping around the engagement engine.
• The model cannot publicly acknowledge its real margins of uncertainty. The creator of a 90% prediction likely understands privately that 50% would be equally defensible. But admitting that would collapse the entire premise that makes people return daily for updated odds.
• The system breaks if anyone asks the obvious question. If the honest answer is "we think one candidate wins but we're genuinely uncertain," the natural follow-up becomes "then why does any of this exist?" That question threatens the entire business model.